Cognitive computing is a generic term describing the combination of diverse ‘modern’ techniques; AI, NLP, machine learning, image analysis and more. You can also see it as a path towards general AI (in contrast to narrow AI). The term ‘augmented intelligence’ usually refers to assisting a person perform some task more efficiently. It often means mimicking the mundane and boring so the actual smart and intuitive decision process remains in the hands of the professional.
An intelligence officer is a person employed by an organization to collect, compile and/or analyze information which is of use to that organization. Emergency calls, anti-terorism organizations typically employ people trained to gather info from various sources. One typically divides these sources in the following domains:
- OSINT or open source intelligence: anything you can find on Wikipedia, newspapers and so on
- IMINT or imagery intelligence: visuals in all shapes and forms together with image analysis (say, detecting salient features and face detection)
- FININT or financial intelligence: information about the financial affairs of entities of interest
- TECHINT or technical intelligence: factual info about weapons, devices and whatnot
- COMINT or communication intelligence: the actual input from a call, signals from sensors, wiretapping and so on
- HUMINT or human intelligence: info from partners, private or internal exchanges
Augmented intelligence systems try to consolidate and automate the process of gathering info. A lot of tasks can nowadays be automated:
- collecting data from social networks, highlighting keywords and using machine learning to classify it
- harvesting data from open databases (tax registers, WolframAlpha…)
- using vision API’s to manipulated or extract info from images
- using language processing to classify sentiments, detect tokens of interest
There is an overwhelming amount of data about everything and everyone out there. An augmented intelligence gathers this data, analyzes it and stores it for further processing:
- AI and machine learning pipeline are used per data source
- the diversity of data types means that one needs to use a combination of relational stores, semantic stores (Apache Jena, StarDog and such) and big data stores (say, Apache Hive)
- neural networks are often complemented with rule-based systems either to enforce a particular emphasis or because there is not enough data to train a network
- a lot of proprietary software and techniques are used you will never read about and well-hidden behind (fire)walls
Because of the diversity and the amount of data many new data visualization technique are used to present things concisely. So, also on the UI level things can be challenging to create compelling and accurate dashboards. Thanks to WebGL and GPU accelerated frameworks one can nowadays create startling viz.
Much of my work as a consultant in the past 15 years has revolved around the domains highlight above and comes together in cognitive computing and augmented intelligence. While I cannot delve much into the details of the projects and customers the following bits can give you an idea:
- consolidating data from diverse sources always boils down to the creation of graphs. It used to be client-side graphs while these days the focus is on massively large graphs server-side. Things like Apache GraphFrames reaching billion of edges or Neo4J stores with a trillion nodes is now accepted. Graph layout and dataviz are just as important as before except that smart pre-processing and filtering is also part of the job.
- AI and neural networks used to be some secret sauce but are now everywhere. Creating neural nets is however often the end-result of some data-lake development and data management sub-projects. The boundary between big-data and AI is where many companies struggle.
- a lot of intelligence happens where it’s difficult to do business: Middle-East, governments, Israel and so on. The good is that there is usually no shortage of money or technology, the bad is that hierarchy, secrecy and rules are killing.
- you need to be technology-independent and technology-proficient at the same time. Customers and solutions decide the technology. You need to understand many domains. Something I find in fact very enjoyable.
- speaking multiple languages is an advantage but it’s even more important to understand multiple cultures, affinities and sensitivities
- no matter how much time and energy you put into smart stuff (i.e AI, ML and all that) the compelling factor is always the dataviz and the visuals. People (men in particular) are most of the time driven by visuals and simple representations.
- forensic analysis and the post-mortem business is more narrow than augmented intelligence. Less challenging but involves many of the techniques.
From a scientist point of view I’m always surprised to see that intelligent software is swallowing more and more mathematics. When quantum computing will take the main stage this will engender a new era of computational intelligence. Number theory (through cryptography), topology (graph layout, topological data analysis) … it has become the new normal. It wasn’t like this twenty years ago. This is great but it also enlarges the gap between who understands and who doesn’t.
As a consultant I see lots of big consultancy firms enjoying the new age but you’ll find very few who can embrace the many aspects of cognitive intelligence. Innovation is not just-another-service-in-the-cloud. Sometimes you need a different type of professional to go out in the jungle and trace a new road.